The Kaiju Who Couldn't Care Less

Riku is my Munchkin cat. In the series he plays Rikugon, a kaiju the size of a building who never does anything on purpose. A Defense Force treats him as an extinction-level threat, and a 1970s news crew covers it live. He wants somewhere warm to sleep.

The engine is the distance between how seriously everyone takes him and how little he cares. The fights borrow from video games: name cards, combo counters, a result screen. You know the cat wins, so you stay to see how. Every episode ends back on the sofa, with Riku waking up and one object from the dream still in the room. The series never says it was a dream.

Riku is a real Munchkin, short legs and all. These photos and this clip come from my own files, and they are the only real footage of him on this page.

Three of the stills are among the four photos I gave the image model to build his character sheet. The sheet locks his face and coat, so every generated shot starts from the same cat. The real-cat scenes at the end of each episode are generated too.

Every episode gives Rikugon a want a real cat would have. The Defense Force reads it as an attack, a game-style fight follows, and the real Riku wakes up with one object from the dream. Each episode changes the opponent and the joke while the engine stays put.

Episodes 1 to 3 are out, and each one plays from its card below. Episodes 4 and 5 are next, built on the same template.

I started with four looks on mood boards: 1970s tokusatsu (特撮, Japanese practical-effects film), black-and-white newsreel, 1987 TV news and a miniature diorama. The first lock on 10 October used the TV-news look as the frame, with the Defense Force footage in tokusatsu. My v2 notes the same day moved everything, news segments and game HUD included, into the 16mm tokusatsu look.

The look kept moving at the shot level. Episode 5 first came out clean and glossy. On 16 October I asked for the original miniature-film look back, with the layout and staging unchanged. A flat overcast sky and a grimy rooftop made the same shot read as a model on a stage.

Locked character sheets come first, and a character without a sheet doesn't go on screen. From the sheets I make a clean still with no text in it. The still becomes video in Seedance. After Effects takes the clip, grades it, and adds what the models handle badly: captions, the game HUD and all the type.

Two Seedance models split the work. The 2.5 model takes key shots, anything where Riku's face and coat have to read, and the beat an episode hangs on. The faster 2.0 model takes wides, cutaways and drafts. Footage is generated at 480p and upscaled in After Effects under film grain, which hides the resolution. Text goes on at full resolution, so it stays sharp.

Here is one Episode 1 shot through every stage: the Defense Force robot lifting off. The board draws it above a hangar. The clean still moves the launch to a rooftop pad. Seedance turns the still into the raw clip, and After Effects grades it and puts the speaker label and subtitle on top.

The clip drifts from the still as it plays, so I matched the two by shot number and by the set. The rooftop lights, the hazard stripes and the ladder carry across.

Nothing is generated from a loose idea. Each episode gets a board first, with every shot timecoded and given its spoken line, its sound and a tag for how it will be made. A shot is either an existing clip with exact in and out points, a start frame for a new generation, or a new shot with a written prompt.

The tags stop me from regenerating footage I already have. Red notes on a frame mark a known flaw, like a crow's net that is missing from every take, or a cut that has to end before a muzzle flash.

Generated characters only work if they stay the same character. Each one has a contact sheet: the kaiju, the Defense Force, the news crew, two robots and a crow. Every shot carries the sheet of everyone in it.

The look has rules too. Heavy grain, faded warm color, halation, plaster miniatures, painted skies. The enemy robot is deliberately not Gundam. The theme is an original march with a straight-faced cat chorus. The roar is Riku's real meow, pitched down one to two octaves under a hall reverb.

AI video gives me footage. An episode takes timing, grade, sound and a layer of graphics, and those get built in After Effects.

Captions, the KAIJU-SCAN threat meter, name cards and the result screen are all drawn in the edit, in exact type. The models never generate them. The screenshots below show Episode 1 in the edit.

The models make a lot of footage fast. Most of my time goes into deciding which of it to use, and into what happens when none of it works.

Around the episodes sit five side stories, each in a different TV format with its own palette and its own picture idea. They are boarded like the episodes, then shot in chains: every clip starts from the last clip's final frame, so a format needs only a handful of new pictures.

The formats are borrowed on purpose. A kaiju who never cares is funniest when he turns up inside things that are supposed to be glamorous.

The series promo is boarded as a trailer and short spots. Every frame on those boards carries a tag. Where footage exists, the board gives exact in and out points. A new shot gets requested only when nothing on disk will do.

Reuse keeps the look consistent as well as cheap.

These are the dated decisions in order, taken from my project notes. The last row rests on file dates instead of a note.

Episode 1 is the template the other episodes are built from, and the series carries a "by NodeMode Studio" credit.

It also works as a test bed for the studio's method: locked references, board-first shooting and a scripted After Effects build with a hand edit on top.